Dr Jeff Dalton

Chuan Meng, Litu Ou, Sean MacAvaney et al.

Deep research has emerged as an important task that aims to address hard queries through extensive open-web exploration. To tackle it, most prior work equips large language model (LLM)-based agents with opaque web search APIs, enabling agents to iteratively issue search queries, retrieve external evidence, and reason over it. Despite search's essential role in deep research, black-box web search APIs hinder systematic analysis of search components, leaving the behaviour of established text ranking methods in deep research largely unclear. To fill this gap, we reproduce a selection of key findings and best practices for IR text ranking methods in the deep research setting. In particular, we examine their effectiveness from three perspectives: (i) retrieval units (documents vs. passages), (ii) pipeline configurations (different retrievers, re-rankers, and re-ranking depths), and (iii) query characteristics (the mismatch between agent-issued queries and the training queries of text rankers). We perform experiments on BrowseComp-Plus, a deep research dataset with a fixed corpus, evaluating 2 open-source agents, 5 retrievers, and 3 re-rankers across diverse setups. We find that agent-issued queries typically follow web-search-style syntax (e.g., quoted exact matches), favouring lexical, learned sparse, and multi-vector retrievers; passage-level units are more efficient under limited context windows, and avoid the difficulties of document length normalisation in lexical retrieval; re-ranking is highly effective; translating agent-issued queries into natural-language questions significantly bridges the query mismatch.

Yola Jones, Fani Deligianni, Jeff Dalton et al.

Worldwide, many millions of people die suddenly and unexpectedly each year, either with or without a prior history of cardiovascular disease. Such events are sparse (once in a lifetime), many victims will not have had prior investigations for cardiac disease and many different definitions of sudden death exist. Accordingly, sudden death is hard to predict. This analysis used NHS Electronic Health Records (EHRs) for people aged $\geq$50 years living in the Greater Glasgow and Clyde (GG\&C) region in 2010 (n = 380,000) to try to overcome these challenges. We investigated whether medical history, blood tests, prescription of medicines, and hospitalisations might, in combination, predict a heightened risk of sudden death. We compared the performance of models trained to predict either sudden death or all-cause mortality. We built six models for each outcome of interest: three taken from state-of-the-art research (BEHRT, Deepr and Deep Patient), and three of our own creation. We trained these using two different data representations: a language-based representation, and a sparse temporal matrix. We used global interpretability to understand the most important features of each model, and compare how much agreement there was amongst models using Rank Biased Overlap. It is challenging to account for correlated variables without increasing the complexity of the interpretability technique. We overcame this by clustering features into groups and comparing the most important groups for each model. We found the agreement between models to be much higher when accounting for correlated variables. Our analysis emphasises the challenge of predicting sudden death and emphasises the need for better understanding and interpretation of machine learning models applied to healthcare applications.

Chuan Meng, Jiqun Liu, Mohammad Aliannejadi et al.

Using large language models (LLMs) to predict relevance judgments has shown promising results. Most studies treat this task as a distinct research line, e.g., focusing on prompt design for predicting relevance labels given a query and passage. However, predicting relevance judgments is essentially a form of relevance prediction, a problem extensively studied in tasks such as re-ranking. Despite this potential overlap, little research has explored reusing or adapting established re-ranking methods to predict relevance judgments, leading to potential resource waste and redundant development. To bridge this gap, we reproduce re-rankers in a re-ranker-as-relevance-judge setup. We design two adaptation strategies: (i) using binary tokens (e.g., "true" and "false") generated by a re-ranker as direct judgments, and (ii) converting continuous re-ranking scores into binary labels via thresholding. We perform extensive experiments on TREC-DL 2019 to 2023 with 8 re-rankers from 3 families, ranging from 220M to 32B, and analyse the evaluation bias exhibited by re-ranker-based judges. Results show that re-ranker-based relevance judges, under both strategies, can outperform UMBRELA, a state-of-the-art LLM-based relevance judge, in around 40% to 50% of the cases; they also exhibit strong self-preference towards their own and same-family re-rankers, as well as cross-family bias.